Navjot Kukreja

> it says that doing solver.op() multiple times increase mempry consumption It doesn't?

The PR https://github.com/opesci/devito/pull/427 was also trying to highlight the same issue but I closed it as duplicate. This is now the blocker for https://github.com/opesci/devito/pull/397 . I have added a test...

I think the current evidence points at the fork before compilation being the culprit.

I have linked the relevant test in my original message - it's the FWI gradient test. Here's the output, if it helps: ``` FAILED tests/test_gradient.py::TestGradient::test_gradientFWI[True-shape0-OT2-16] - assert False FAILED tests/test_gradient.py::TestGradient::test_gradientFWI[False-shape0-OT2-16]...

Can we break this down further into TODOs? PS: I don't understand why I didn't get a notification for this tag.

Yeah we probably don't want to do `abs(float)` as that would just generate the reverse warning. Can we infer types in some cases though? For example, in this case, I...

You're saying two things here: 1. first touch wasn't updated to use the new API 2. Do we need it? Interestingly I discovered this around the same time that you...

The following code snippet was tested on Richter: ``` from time import time from devito import TimeData u = TimeData(name='u', shape=(1000, 200, 200, 200)) start = time() u.data[:]=1 end =...

And now on the `first_touch_again` branch, this snippet: ``` from time import time from devito import TimeData u = TimeData(name='u', shape=(1000, 200, 200, 200), first_touch=True) start = time() u.data[:]=1 end...

With #332 in master, this test can be carried out and reproduced by everyone here on master. Let's continue this conversation? @FabioLuporini @mlange05